Double Trouble: How to not Explain a Text Classifier’s Decisions Using Counterfactuals Synthesized by Masked Language Models? (2022.aacl-main)
Copied to clipboard
| Challenge: | Input Marginalization (IM) is a method that takes the prediction difference between before-and-after an input feature (here, a token) is removed as its attribution. |
| Approach: | They propose to use a BERT-based method to replace a token with a feature to give more plausible counterfactuals. |
| Outcome: | The proposed method is effective, but the Deletion-BERT metric is biased towards IM, and the results are not convincing. |
Similar Papers
Interpretation of NLP models through input marginalization (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to interpret NLP predictions replace each token with a predefined value, resulting in misleading interpretations. |
| Approach: | They propose to marginalize each token out of the training data distribution to demystify the "black box" property of deep neural networks for natural language processing. |
| Outcome: | The proposed method marginalizes each token out of the training data distribution. |
Linear Classifier: An Often-Forgotten Baseline for Text Classification (2023.acl-short)
Copied to clipboard
| Challenge: | Large-scale pre-trained language models such as BERT are popular solutions for text classification. |
| Approach: | They argue that large-scale pre-trained language models such as BERT are popular solutions for text classification . authors argue that running a simple baseline like linear classifiers on bag-of-words features is important for text classification . |
| Outcome: | The proposed approach may only sometimes get satisfactory results for some problems. |
Exploring the Role of BERT Token Representations to Explain Sentence Probing Results (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have focused on enhancing existing models with the primary objective of improving downstream performance on various NLP tasks. |
| Approach: | They propose to use BERT to encode meaningful knowledge in token representations to explain probing results. |
| Outcome: | The proposed model can detect syntactic and semantic abnormalities and distinguish between grammatical number and tense subspaces. |
From BERT‘s Point of View: Revealing the Prevailing Contextual Differences (2022.findings-acl)
Copied to clipboard
| Challenge: | BERTology is a new approach to understanding the inner workings of large pretraining language models. |
| Approach: | They propose to invert the probing design to analyze the prevailing differences and clusters in BERT’s high dimensional space by extracting coarse features from masked token representations and predicting them by probing models with access to only partial information. |
| Outcome: | The proposed method extracts coarse features from masked token representations and predicts them by probing models with access to only partial information. |
Instance-Selection-Inspired Undersampling Strategies for Bias Reduction in Small and Large Language Models for Binary Text Classification (2025.acl-long)
Copied to clipboard
Guilherme Fonseca, Washington Cunha, Gabriel Prenassi, Marcos André Gonçalves, Leonardo Chaves Dutra Da Rocha
| Challenge: | Existing methods to mitigate class imbalanced datasets are limited by existing methods. |
| Approach: | They propose two undersampling methods inspired by state-of-the-art Instance Selection techniques to mitigate class imbalance bias in ATC. |
| Outcome: | The proposed methods reduce classifier bias (56%) across all datasets without effectiveness loss while improving efficiency (1.6x speedup), scalability and reducing carbon emissions (up to 50%). |
“Will You Find These Shortcuts?” A Protocol for Evaluating the Faithfulness of Input Salience Methods for Text Classification (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing work on faithfulness evaluation is not conclusive and does not provide a clear answer as to how different methods are to be compared. |
| Approach: | They propose a protocol for faithfulness evaluation that makes use of partially synthetic data to obtain ground truth for feature importance ranking. |
| Outcome: | The proposed method is based on partially synthetic data and is compared with lexical shortcuts on a range of datasets and LSTM models. |
Revisiting Token Dropping Strategy in Efficient BERT Pretraining (2023.acl-long)
Copied to clipboard
| Challenge: | Token dropping is a recently-proposed strategy to speed up the pretraining of masked language models, such as BERT. |
| Approach: | They propose a semantic-consistent learning method to improve token dropping by skipping the computation of a subset of input tokens at several middle layers. |
| Outcome: | The proposed method achieves consistent and significant performance gains across all tasks and model sizes. |
Extracted BERT Model Leaks More Information than You Think! (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing pre-trained language models are vulnerable to model extraction attacks . model extraction can cause severe privacy leakage even when victim models are facilitated with state-of-the-art defensive strategies. |
| Approach: | They propose to launch an attribute-inference attack against an extracted BERT model to prevent privacy leakage. |
| Outcome: | The proposed attack can cause severe privacy leakage even when victim models are facilitated with state-of-the-art defensive strategies. |
How Far Is Too Far? Studying the Effects of Domain Discrepancy on Masked Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Pre-trained masked language models perform strongly on a wide variety of NLP tasks. |
| Approach: | They propose a mechanism to quantify the difference in domains between the pre-trained model and the task and partition it using a cloze task. |
| Outcome: | The proposed model performs better on openly available e-commerce datasets than the original model on scientific and biomedical datasets. |
Should you marginalize over possible tokenizations? (2023.acl-short)
Copied to clipboard
| Challenge: | Autoregressive language models map token sequences to probabilities. |
| Approach: | They propose to marginalize over all tokenizations to compute the probability of a string . they find that the gap in log-likelihood is no larger than 0.5% in most cases . |
| Outcome: | The proposed algorithm shows that the gap in log-likelihood is no larger than 0.5% in most cases . but it becomes more pronounced for data with long complex words, the authors show . |